The behavior of networks under static conditions is well-understood. However, under dynamic conditions, when routes change, network behavior can become indeterminate, and a variety of pathological conditions can occur. In a standard network, for example, where route updates occur at each switch independently and asynchronously, in response to a variety of network events, the behavior of the network at any time is only stochastically predictable.
Some networks such as implemented by OpenFlow implement controllers to choose routes. For example, in an OpenFlow network, routes are chosen by a controller, and not by the action of an independent algorithm running on an individual switch. Thus, updates can occur at times chosen by the controller. But even in such networks, the controller cannot implement all switch updates instantaneously and synchronously. Switch updates can take and time, and are realized at different and unpredictable for each switch in the network. As a result, the path taken by a packet is dependent on which switches were updated and in which order. Further, packet arrival times at individual switches are unpredictable.
Embodiments described herein implement a transition protocol that can be implemented by a controller of a network. More specifically, a protocol is implemented on a network that provides for each packet passing through a network to be consistently handled by a single set of rules, even dynamic conditions when routing rules of the network are changed.
Among other benefits, implementation of a protocol such as described enables switch resources to be conserved, ensuring that only a single set of rules is present on a switch at any time. In examples described herein, when a network changes from one rule set (Rule Set 1) to another (Rule Set 2), the packets affected by the change are computed, and packets affected by the change are sent to the controller. When each switch has been updated to send affected packets to the controller, Rule Set 2 is sent to the switches and packets sent to the controller are rereleased into the network.
According to examples described herein, a controller for a network is provided. The controller is configured to implement a transition protocol in which a first rule set for routing packets received by a group of switches during a first time period is to be updated to a second rule set. The controller implements the transition protocol by implementing, during a first transition period, a transition rule set. The transition rule set provides that (i) any switch in the group of switches that is to forward packets to a same given node under both the first rule set and the second rule set will continue to forward packets to the given node during the first transition period, and (ii) any switch in the group of switches that is to forward packets to a node under the second rule set that is different than a node that the switch forwarded packets to under the first rule set will forward packets to the controller. During a subsequent time period following the first transition period, packets received from any switch in the first transition period are sent by the controller to a destination node in accordance with the second rule set, and implementing the second rule set for each switch in the group of switches.
In variations, a group of switches can be operated by each switch individually forwarding each packet to a next hop for that switch as designated by a first rule set. When a transition rule set is received from a controller, individual switches in a first set of switches forward each packet to the next hop for that switch as designated by both the first rule set and by a second rule set that is to be implemented after the first rule set. During a transition period after the transition rule set is received, a second set of switches forward each packet to the controller for forwarding to a destination node that is to be designated by the second rule set. The switches of the second set may include those switches that are to have their next hops changed as a result of the change from the first rule set to the second rule set.
Still further, a transition protocol is provided herein in which a first rule set for routing packets received by a group of switches during a first time period is to be updated to a second rule set. During a transition period, at least some switches in the group of switches route packets to a controller, while other switches in the group of switches route packets to a next hop that is unchanged by the change in the rule set. The controller forwards packets that are received from at least some of the switches in the group to a destination node each of the packets, as determined from the updated rule set.
Controller
FIG. 1 illustrates an example of a controller for a network. In an example of FIG. 1, a network 10 includes a controller 100, edge switches 12, and a group of switches 20. The controller 100 includes logic 110 that implements rule sets to control how each switch forwards a data packet. In the example of FIG. 1, the controller can implement a first rule set 111, then transition to a second or updated rule set 113. In order to transition between rule sets 111, 113, controller 100 includes transition logic 120 that implements a transition protocol. The transition protocol includes determination of a set of transition rules 121 that are determined by the transition logic 120. The transition logic 120 can determine the transition rules 121 based on a determination of packets that would be treated differently under the second rule set 113. For those packets, the transition logic 120 dictates that the packets are forwarded to the controller 120. As described with examples below, the transition rules 121 ensure that individual packets received by each of the switches 20 are handled consistently by a single rule set, even during an update between rule sets.
As mentioned, when implemented, the transition rules 121 may require some switches 20 in the group to forward packets to the controller 100. Packet forwarding logic 130 on the controller 100 can forward packets to a destination node (e.g., another switch, edge node). The packet forwarding operation performed by the controller 100 can insert the packet at any point in the packet's path that is between the forwarding switch 20 and the packet's destination node within the packet, in accordance with the second rule set 113. For example, the controller 100 can insert the packet at a next hop for the forwarding packet or at some other node prior to the destination node for the packet.
In an example of FIG. 1, the transition protocol can also implement timing stages (or epochs) during which transition between rules occurs on each switch. Among other determinations, the timing logic 140 can determine when the controller 100 signals the updated or second rule set 113 after signaling the transition logic 121. The timing logic 140 can extend the duration for when the transition rules 121 is in effect based on each switch 20 acknowledging receipt of the transition rules 121. Additionally, the timing logic 140 can also account for network latency, so as to maintain the transition rules 121 for duration of time after receipt of acknowledgements from each switch 20, where the duration is based on a latency measurement that is as long as the maximum estimated time for a packet to transition the network.
Methodology
FIG. 2 illustrates an example method for implementing a transition protocol to control switches of a network in updating a rule set for routing packets on the network. In one implementation, a method such as described by an example of FIG. 2 can be implemented in an OpenFlow network. In describing an example of FIG. 2, reference may be made to components of FIG. 1 for purpose of illustrating suitable elements for performing a step or sub-step being described.
The controller 100 can update a rule set for routing packets on each switch (210). For example, the controller 100 can implement a first rule set F(1) during a time period, then implement a sequence of operations to implement a second rule set F(2) at a subsequent time period. A transition protocol may be utilized in updating from rule set F(1) to rule set F(2). The transition protocol may be implemented with multiple operations to ensure individual switches consistently utilize a single set of rules in routing packets on the network 10.
In an example of FIG. 2, the switches 20 in the network that are to maintain their next hop to be the same after the updated rule set F(2) is implemented continue routing to their respective next hop (212). The switches that are to have their next hop changed as a result of the second rule set F(2) route their data packets to the controller 100 (214).
Subsequently, the controller forwards data packets it receives from the switches 20 to destination nodes in accordance with the updated rule set F(2) (216). The controller 100 does not necessarily need to send packets to next hops for those switches 20 that forwarded packets. Rather, the controller 100 can forward the packets to a node (e.g., downstream node) along the route of the data packet to ensure the data packet is received at its final destination.
Subsequently, the controller 100 implements the second rule set for the switches 20 (220). After a duration in which the transition rules 121 are signaled to the switches 20, the second rule set F(2) can be signaled to the switches 20.
The timing in which the controller 100 implements the second rule set F(2) can be set by a duration that is determined by timing logic 140. In particular, the timing logic 140 determine a transition time that corresponds to each switch acknowledging the transition rule set 121, followed by an interval that corresponds to a measurement of network latency (e.g., the estimated maximum time for a data packet to traverse the network).
In examples described by FIG. 1 and FIG. 2, each rule set may be implemented as a transfer function. For a packet p, it is assumed there is a network transfer function F(p). For purpose of description below, the transfer function F1(p) represents the Rule Set 1, the transfer function F2(p) represents the Rule Set 2, and the transfer function F12(p) represents the transition rules between Rule Set 1 and 2.
With further reference to examples of FIG. 1 and FIG. 2, an embodiment recognizes that a given switch cannot anticipate if other switches 20 in the network are forwarding packets according to Rule Set 1 or Rule Set 2. Furthermore, an embodiment recognizes that it is unsafe for any switch 20 to forward a packet which will be handled differently under different rule implementations provided from the controller 100. However, an embodiment also recognizes that it is safe for the data packets to be sent to the controller 100. Conversely, if a switch 100 will not modify a packet under the Rule Set 2, the switch can send the data packet on.
Each switch therefore sends to the controller those packets it will handle differently under Rule Set 2 and Rule Set 1. These packets are handled by transition rules which direct the packets to the controller 100. Once the controller 100 receives acknowledgement from each switch that it has received the transition rules, it can determine when it is safe to determine that no switches are forwarding packets in accordance with the Rule Set 1. Once this confirmation is made, the controller 100 can send Rule Set 2 to the switches.
According to some embodiments, the controller 100 implements the change to the rule set during four sequential epochs, each measured at the controller 100. In the first epoch, all switches are using Rule Set 1. A transfer function at each switch is fs1, and each packet is processed exclusively by this family of transfer functions. As such, F(p)=Fl(p) for all packets p.
In the second epoch, the transfer functions fs12 are loaded onto each switch. During this epoch, the transfer function at a switch starts at fs1 and ends as fs12. Thus, F(p)=F1(p) or F12(p) for each packet p, depending upon the relative arrival time of fs12 and p at each switch s for which fs1 (p) does not equal fs12 (p).
Each switch sends a completion signal to the controller when a new rule set is loaded. Once the last switch has sent a completion signal to the controller, the controller waits for the maximum network latency.
At this point, controller 100 recognizes epoch three. At the start of epoch three, the controller sends fs2 to all switches, and releases all packets sent to the controller 100 during epoch two. During epoch three, Fp=F12(p) or F2(p) for each packet p, depending upon the relative arrival time of fs2 and p at each switch s. Since a packet p may arrive at a switch s whose transfer function is still fs12, the packet may be sent to the controller 100. However, the controller 100 can continuously releases packets sent to the controller during epoch three, and so these packets will be delivered to their destination. As a result, all packets sent during epoch three have F(p)=F2(p).
Once the last completion signal has been sent to the controller 100 from the switches, the controller 100 waits for the maximum network latency. Epoch four then begins. During Epoch four, fs=fs2 for all switches s, and all packets sent into the network in epoch four are processed entirely by switches using the transfer function fs2; hence F(p)=F2(p) for all packets p. After epoch four begins, a new update can begin at any time. The four epochs are summarized in Table 1.
EpochTransfer FunctionPacket FunctionEnd1f1sF1f12s sent to switches2f1s → f12sF1 → F12One max network latencyafter completion signal3f12s → f2sF12 → F2One max network latencyafter completion signal4f2sF2N/A
Example Network Implementation
FIG. 3A and FIG. 3B illustrate examples for updating rule sets for routing data packets, including implementing a transition protocol when updating the rule sets.
More specifically, FIG. 3A illustrates the network 300 includes edge switches 310, a group of switches 320 that route packets, and a controller 3330. The Rule Set 1 (fs1) is represented by solid lines and illustrates the routing performed by each switch when the first rule set is in implementation. The Rule Set 2 (fs2) is represented by dashed lines and illustrates the routing performed by each switch 320 when the second rule set is in implementation. A transition protocol in accordance with various examples described herein may be implemented in order to enable the network 300 to transition from implementing Rule set 1 to Rule Set 2. The implementation of the transition protocol is illustrated by FIG. 3B.
In the example of FIG. 3B, the network 300 with the group of switches incorporating the transition set of rules (fs12). The transition set of rules are shown in broken rules. During a transition period, some of the switches 320 forward packets to the controller 310. As shown by an example of FIG. 3A and FIG. 3B, switches 320 (u, v) have their respective next hop changed as between Rule Set 1 and Rule Set 2. Accordingly, the switches 320 (u,v) forward packets to the controller 310 during the transition phase, as shown by FIG. 3B.
The other switches 320 (w, y, s, z and t) do not have changes to their respective next hops as a result of the Rule Set 1 being updated by the Rule Set 2. Instead, each of the switches 320 (w, y, s, z and t) forwards packets according to either Rule Set 1 (fs1) or Rule Set 2(fs2).
In the example of FIG. 3B, packets that are communicated from, for example, edge nodes 302 to edge node 304, and for which the routing is changing as a result of the change in rule sets are handled at either switch 320 (u) (in which case the packet is forwarded to the controller 330), or by switch (v) (in which case the packet is routed to the controller 320). Otherwise, the packets were handled by the Rule Set 1 (fs1), and the handling would be the same as for Rule Set 2 (fs2).
With regard to switches w, t, the switches do not receive packets under Rule Set 1 (fs1), and therefore will not handle packets in the transition. These switches can thus forward packets as before.
Controller Hardware
FIG. 4 illustrates an example controller for use with a network, as described with examples of FIG. 1, FIG. 2 and FIG. 3A and FIG. 3B. The controller 400 can be implemented in a variety of forms, including as a dedicated appliance, or through software executed on a server or machine. The controller 400 can include processor 405, memory resources 406, and a communication interface 418. The memory resources 406 can store instructions that enable the processor to implement functionality, including a transition protocol 415 (e.g., transition rules based an update to an existing rule set). The processor 405 can execute the instructions for the transition protocol 415. In implementing the transition protocol 415, the processor 405 can communicate with the communication interface 418 to send routing rules (E.g., transition rules, update rules), as well as to receive and forward transition packets 425.
Although illustrative embodiments have been described in detail herein with reference to the accompanying drawings, variations to specific embodiments and details are encompassed by this disclosure. It is intended that the scope of embodiments described herein be defined by claims and their equivalents. Furthermore, it is contemplated that a particular feature described, either individually or as part of an embodiment, can be combined with other individually described features, or parts of other embodiments. Thus, absence of describing combinations should not preclude the inventor(s) from claiming rights to such combinations.